Interview, Fireside Chat
Can AI Solve Healthcare's Urgent Workforce Challenges? w/ Ankit Jain
- Infinitus deploys a stack of large language models (LLMs) and small language models (SLMs) to address healthcare workforce shortages through autonomous voice and data agents.
- The platform has processed over 5 million phone calls and 100 million hours of machine-to-human audio interactions within a five-year period.
- Initial viability was proven in 2019 (pre-hype) via a demo call to UnitedHealthcare, where a machine successfully verified benefits for a fictional patient ("Bruce Willis") despite being unable to hide its identity as an AI.
- The company utilizes a dual-pillar strategy: fully autonomous agents for end-to-end task completion (e.g., benefit verification) and "co-pilot" tools (e.g., FastTrack) to handle wait times for human employees.
- The "co-pilot" product waits on hold within IVR systems and transfers the call to human staff only when a live agent is ready, eliminating administrative downtime.
- Infinitus mitigates hallucination risks through a "discrete action space" architecture and multiple guardrails (technical, human, and compliance-based) to ensure clinical safety.
- The company reports that human agents on the payer side provide inconsistent answers 25% of the time on the same query due to manual errors, a variance that the AI system detects and corrects via real-time QA.
- Through proactive "push-back" when discrepancies are detected during calls, Infinitus validates and corrects 70–80% of erroneous data points immediately.
- The company sells primarily to operations leaders and P&L owners rather than IT budgets, positioning the solution as an elastic workforce layer to handle peak "blizzard" periods (e.g., January eligibility checks).
- Current deployments have reduced phone call volumes by single-digit percentages but successfully digitized 30–40% of data that previously required synchronous voice conversations.
- Infinitus has successfully encouraged partners to establish APIs for data exchange after demonstrating that their bots can handle the initial friction of manual workflows, though deep document data often still requires LLM extraction before API availability.
- The company avoids "bot-to-bot" conversations where possible, advocating for machine-to-machine data transfer (bits and bytes) rather than English-language dialogue, though enterprise security protocols sometimes force continued voice interfaces.
- Infinitus's competitive moat relies on proprietary data fine-tuning, deep workflow integration (last-mile integration), and the ability to navigate fragmented healthcare SOPs that "one-size-fits-all" models (e.g., Google, Amazon) cannot easily address.
- The company claims to support 44–45% of the Fortune 50, navigating recent enterprise AI review boards and compliance checklists without being perceived as a generic technology purchase.
- Talent acquisition prioritizes passion for healthcare problem-solving over pure technical skill, leveraging a trend of Silicon Valley professionals seeking meaningful impact in health.
- The company observes that patients often lose clarifying questions after diagnosis due to cognitive overload, necessitating proactive, multi-modal (voice/text) education platforms rather than reactive inquiry models.
- The company acknowledges that while APIs are the long-term goal for back-office communication, front-office patient interactions will remain multi-modal (voice, text, chat) to meet diverse patient preferences.
- Future roadmap items include integrating disparate social determinants of health data, which is currently missing in 80% of cases, to enable fully context-aware personalized care.
- Infinitus predicts that within five years, the convergence of fragmented data and advanced LLM translation capabilities will enable care experiences currently impossible due to data scarcity.